pre-action-review

Pre-action review

Before you execute an action that would be hard or impossible to cleanly undo, call the review tool (via the invinoveritas MCP connection) and pass a plain-text description of exactly what you're about to do. It returns an independent approve / approve_with_concerns / reject verdict with a confidence score and, on reject, specific blocking issues.

When to use this

Call it before, not after, whenever the action fits any of these:

• Destructive or hard-to-reverse shell commands — rm -rf, dropping a database/table, force-pushing over history, deleting a cloud resource, revoking credentials.
• Merges and branch operations at scale — this matters more than usual on Origin-style workflows where many agents branch, resolve conflicts, and merge with little or no human in the loop per merge. A signed pre-merge verdict is a real, checkable second opinion on a merge that might otherwise go straight to main unreviewed.
• Production deploys and migrations — anything that changes live infrastructure, schema, or configuration in a way a rollback might not cleanly reverse.
• Financial or irreversible-by-nature actions — a payment, a transfer, a trade, deleting a user's data, sending an external communication on someone's behalf.
• Anything you're not fully confident about — if you'd want a second pair of eyes on this before running it yourself, that's the signal to call review.

Do NOT call it for routine, easily-reversible edits (a normal code change on a feature branch, a local test run, reading a file) — that's noise, not signal, and burns the user's call budget.

How to call it

review(
  artifact="<plain description of the exact action: what command/merge/deploy, on what target,
             with what data at stake>",
  artifact_type="general"   # or "trade" / "onchain_action" for financial actions
)

Be concrete: name the actual command, the actual branch/target, and any stated safeguards (a tested backup, a dry-run result) — a vague artifact gets a vague, less useful verdict.